Guides
SIGN INSIGN UP

Driving style & Safety Score

Suggest Edits

Introduction

This document describes

  1. How and which events are detected by Telematics SDK
  2. How the scores are generated for every single parameter of the trip
  3. How these scores are aggregated into a total score
  4. How the total driver rating is finally generated in a given period

Rating description presents the universal approach, developed by our company based on many years of experience; most of the input variables presented in this document could be adopted depending on your business needs and the focus on the specific characteristics of clients' driving style.

Since 2019 we have moved to the 3rd Generation of the scoring model, which allows distinguishing a context of events and adds penalty points by a level of risk generated by an event.

Scoring variation

  • The existing product solutions are based on different periods of observation of a client- from one month to a constant period of monitoring.

  • Our analysis shows that approximately in 1 month it is possible to determine a driving style of a client even taking into account his possible intentional updating of the driving style. However, for a more precise determination of the annual kilometrage of a car, which required a significantly longer period, the stabilization of scoring model sets is 3-4 months.

  • From the point of view of obtaining the necessary information, a 3-month monitoring period is sufficient for obtaining steady results.

Variation of telematics score during a monitoring period

scoring variation

Gathered data

  • GPS (1hz) 👉 Raw Telematics Data Format
  • Data from a gyroscope, magnetometer, and accelerometer (60hz)
  • Data on whether the device is locked and the screen is on

Events and key concepts

Overall description of collected data and algorithm of identifying

Speeding

  • At each driving moment, Telematics SDK collects the user's speed
  • This data is compared with the existing speed limits provided by our partners (Location platforms: HERE, Google, etc.)
  • We detect in which points there was speeding and then combine detected time intervals into events with a minimum duration of 3 seconds and maximum gap of 5 seconds between them

Variables influencing the rating:

  • Event duration
  • Speed limit
  • Actual Speed

Phone usage (Distracted driving)

  • Phone usage is detected by ML-model, the main input is the data from (2) and (3)
  • We detect at which points there was phone usage and then combine detected time intervals into events with a minimum duration of 3 seconds and a maximum gap of 20 seconds between them

Variables influencing the rating:

  • Event duration
  • Speed

Acceleration

  • User's phone reads data of item (2)
  • Our algorithms continuously adapt phone axes to calculate correct accelerations.
  • The resulting accelerations in the direction of movement are transformed into an event if the driver has accelerated >3 m/s2.
  • We detect at which points there was acceleration and then combine detected time intervals into events with a minimum duration of 0.6 seconds and a maximum gap of 3 seconds between them.

Variables influencing the rating:

  • Event duration
  • Acceleration value
  • Speed

Braking (Deceleration)

User's phone reads data of item (2)
Our algorithms continuously adapt phone axes to calculate correct decelerations
The resulting decelerations in the direction of travel are transformed into an event if the driver has slowed down >3.2 m/s2.
We detect at which points there was deceleration and than combine detected time intervals into events with a minimum duration of 0.6 seconds and maximum gap of 3 seconds between them.

Variables influencing the rating:

  • Event duration
  • Deceleration value
  • Speed

Cornering

  • User's phone reads data of item (2).
  • Our algorithms continuously adapt phone axes to calculate correct accelerations.
  • The resulting accelerations in the direction of motion are transformed into an event if the perpendicular acceleration to the direction of motion >4.2 m/s2.
  • We detect at which points there was acceleration and then combine detected time intervals into events with a minimum duration of 0.6 seconds and a maximum gap of 3 seconds between them.

Variables influencing the rating:

  • Event duration
  • Deceleration value
  • Speed

Trip scoring

This section describes the way of how events and their characteristics are transformed to track rating.

  • Each track has a collection of events
  • Each event has its own characteristics
  • Number of penalty points is proportional to the risk level of each event
  • Points of track:
    • Phone usage points
    • Cornering points
    • Braking point
    • Acceleration point
    • Speeding points
  • Total points of the ride is a weighted sum of parameters' points.
  • Total points of the ride are weighted by the distance of the ride
  • Total and parameter weighted points are transformed to 100-scale
  • Star rating:
    • 100 - 5 star
    • => 90 - 4 star
    • => 80 - 3 star
    • => 70 - 2 star
    • < 70 - 1 star

Overall scoring

This section describes the way how the user receives his aggregate scoring (usually, 2-weeks)

User’s sum of points for 14-day period:

  • Phone usage points
  • Cornering points
  • Braking points
  • Acceleration points
  • Overspeeding points
  • Total points
  • Points of the ride are weighted by the distance of the ride
  • Total and parameter weighted points are transformed to 100-scale

Updated about 1 year ago